Anatomy-driven augmentation of medical images

Abstract

Systems/techniques that facilitate anatomy-driven augmentation of medical images are provided. In various embodiments, a system can access a medical image and a ground-truth segmentation mask corresponding to the medical image, wherein the ground-truth segmentation mask can indicate a location of a first anatomical structure depicted in the medical image. In various aspects, the system can create an augmented version of the medical image and an augmented version of the ground-truth segmentation mask, by applying a continuous deformation field to fewer than all pixels or voxels in the medical image and in the ground-truth segmentation mask. In various instances, the continuous deformation field can encompass: pixels or voxels that correspond to the first anatomical structure; and pixels or voxels that correspond to a surrounding periphery of the first anatomical structure.

What is claimed is: 
     
         1 . A system, comprising:
 a memory configured to store computer executable components; and   a processor that executes at least one of the computer-executable components that:
 accesses a medical image and a ground-truth segmentation mask corresponding to the medical image, wherein the ground-truth segmentation mask indicates respective locations of a first plurality of anatomical structures depicted in the medical image; and 
 creates an augmented version of the medical image and an augmented version of the ground-truth segmentation mask, by applying a continuous deformation field to fewer than all pixels or voxels in the medical image and in the ground-truth segmentation mask, wherein the continuous deformation field encompasses:
 first pixels or first voxels in the medical image that correspond to a first anatomical structure of the plurality of anatomical structures; and 
 second pixels or second voxels in the medical image that correspond to a surrounding periphery of the first anatomical structure, wherein at least a portion of a second anatomical structure of the plurality of anatomical structures is depicted by the second pixels or second voxels that correspond to the surrounding periphery, such that the continuous deformation field deforms the portion of the second anatomical structure commensurately with the first anatomical structure, and wherein deformation of the first anatomical structure is constrained based on a defined criterion associated with the second anatomical structure for retaining biological plausibility of the deformation of the first anatomical structure. 
 
   
     
     
         2 . The system of  claim 1 , wherein the biological plausibility of the deformation of the first anatomical structure is based on a physical limitation associated with the deformation of the portion of the second anatomical structure. 
     
     
         3 . The system of  claim 1 , wherein the medical image is a magnetic resonance imaging scan of a brain, wherein the first anatomical structure is a brain tumor, and wherein the second anatomical structure is a corpus callosum of the brain. 
     
     
         4 . The system of  claim 1 , wherein the medical image is a trans-vaginal ultrasound scan, wherein the first anatomical structure is a fibroid, and wherein the second anatomical structure is an endometrium. 
     
     
         5 . The system of  claim 1 , wherein the at least one of the computer-executable components further:
 trains a deep learning neural network on the augmented version of the medical image and the augmented version of the ground-truth segmentation mask.   
     
     
         6 . The system of  claim 1 , wherein the continuous deformation field tapers to zero at its boundary. 
     
     
         7 . The system of  claim 1 , wherein the continuous deformation field causes the first anatomical structure or the surrounding periphery of the first anatomical structure to expand, contract, translate, rotate, or shear. 
     
     
         8 . A computer-implemented method, comprising:
 accessing, by a device operatively coupled to a processor, a medical image and a ground-truth segmentation mask corresponding to the medical image, wherein the ground-truth segmentation mask indicates respective locations of a plurality of anatomical structures depicted in the medical image; and   creating, by the device system, an augmented version of the medical image and an augmented version of the ground-truth segmentation mask, by applying a continuous deformation field to fewer than all pixels or voxels in the medical image and in the ground-truth segmentation mask, wherein the continuous deformation field encompasses:
 first pixels or first voxels in the medical image that correspond to a first anatomical structure of the plurality of anatomical structures; and 
 second pixels or second voxels in the medical image that correspond to a surrounding periphery of the first anatomical structure, wherein at least a portion of a second anatomical structure of the plurality of anatomical structures is depicted by the second pixels or second voxels that correspond to the surrounding periphery, such that the continuous deformation field deforms the portion of the second anatomical structure commensurately with the first anatomical structure, and wherein deformation of the first anatomical structure is constrained based on a defined criterion associated with the second anatomical structure for retaining biological plausibility of the deformation of the first anatomical structure. 
   
     
     
         9 . The computer-implemented method of  claim 8 , wherein the biological plausibility of the deformation of the first anatomical structure is based on a physical limitation associated with the deformation of the portion of the second anatomical structure. 
     
     
         10 . The computer-implemented method of  claim 8 , wherein the medical image is a magnetic resonance imaging scan of a brain, wherein the first anatomical structure is a brain tumor, and wherein the second anatomical structure is a corpus callosum of the brain. 
     
     
         11 . The computer-implemented method of  claim 8 , wherein the medical image is a trans-vaginal ultrasound scan, wherein the first anatomical structure is a fibroid, and wherein the second anatomical structure is an endometrium. 
     
     
         12 . The computer-implemented method of  claim 8 , further comprising:
 training, by the system, a deep learning neural network on the augmented version of the medical image and the augmented version of the ground-truth segmentation mask.   
     
     
         13 . The computer-implemented method of  claim 8 , wherein the continuous deformation field tapers to zero at its boundary. 
     
     
         14 . The computer-implemented method of  claim 8 , wherein the continuous deformation field causes the first anatomical structure or the surrounding periphery of the first anatomical structure to expand, contract, translate, rotate, or shear. 
     
     
         15 . A computer program product for facilitating anatomy-driven augmentation of medical images, the computer program product comprising a non-transitory computer-readable memory having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:
 access a medical image and a ground-truth segmentation mask corresponding to the medical image, wherein the ground-truth segmentation mask indicates respective locations of a plurality of anatomical structures depicted in the medical image; and   create an augmented version of the medical image and an augmented version of the ground-truth segmentation mask, by applying a continuous deformation field to fewer than all pixels or voxels in the medical image and in the ground-truth segmentation mask, wherein the continuous deformation field encompasses deforms only:
 first pixels or first voxels in the medical image that correspond to first anatomical structure of the plurality of anatomical structures; and 
 second pixels or second voxels in the medical image that correspond to a surrounding periphery of the first structure, wherein at least a portion of a second anatomical structure of the plurality of anatomical structures is depicted by the second pixels or second voxels that correspond to the surrounding periphery, such that the continuous deformation field deforms the portion of the second anatomical structure commensurately with the first anatomical structure, and wherein deformation of the first anatomical structure is constrained based on a defined criterion associated with the second anatomical structure for retaining biological plausibility of the deformation of the first anatomical structure. 
   
     
     
         16 . The computer program product of  claim 15 , wherein the biological plausibility of the deformation of the first anatomical structure is based on a physical limitation associated with the deformation of the portion of the second anatomical structure. 
     
     
         17 . The computer program product of  claim 15 , wherein the medical image is a magnetic resonance imaging scan of a brain, wherein the first anatomical structure is a brain tumor, and wherein the second anatomical structure is a corpus callosum of the brain. 
     
     
         18 . The computer program product of  claim 15 , wherein the medical image is a trans-vaginal ultrasound scan, wherein the first anatomical structure is a fibroid, and wherein the second anatomical structure is an endometrium. 
     
     
         19 . The computer program product of  claim 15 , wherein the program instructions are further executable by the processor to cause the processor to:
 train a deep learning neural network, using the augmented version of the medical image as a training input, and using the augmented version of the segmentation mask as a ground-truth annotation.   
     
     
         20 . The computer program product of  claim 15 , wherein the continuous deformation field tapers to zero at its boundary.